A developer has released KiviDB, an open-source, multi-threaded Redis and Valkey alternative written in Rust. The project began in February 2026 as a way to learn Rust and to address memory blow-ups after snapshots and out-of-memory errors on running pods, issues the author faced with Dragonfly. KiviDB supports both RESP2 and RESP3 protocols, so existing Redis clients work unchanged. Its keyspace is sharded across threads, and it performs best on Linux thanks to io_uring via the tokio_uring crate, with a Tokio fallback on other operating systems. The core engine is licensed under AGPLv3, and the author notes that approximately 98-99% of Redis 8.10 commands are supported.
The author ran benchmarks on two c7gn.metal servers in the same us-east1 placement group using memtier-benchmark, with 10M keys of 100B each across 32 cores. With pipelining at depth 16, KiviDB reached 21.2M GET operations per second, compared with 9.1M for Dragonfly 2.0, 2.4M for Redis 8.10, and 4.7M for a 32-shard Redis Cluster. Without pipelining, Dragonfly was slightly faster at 3.35M ops/sec versus 3.07M for KiviDB. Memory use per key was 177B for KiviDB, 134B for Dragonfly, and 178B for Redis. Vector ingestion was slower than alternatives, while vector queries were faster at lower ef settings but slower than Dragonfly at high ef.
The project also includes built-in full-text and vector search, and the author has built a cloud managed service concept across AWS, GCP and Azure. Known command gaps are tracked on GitHub, and contributors are invited. The project can be tried with a single Docker command, which starts the server on port 6380.
Source: Hacker News · Summarized by HeadlinesBriefing